# AI sprint retrospective analysis — editable worksheet

Blank editorial worksheet, version RETRO-073-v1, revised 2026-09-04. This is not an audit or a required Scrum form. Copy it into an approved workspace and replace blank fields with your own permitted records. Do not upload confidential material to an unapproved tool. The linked article's numerical example is fictional; it does not supply your team's baseline.

## 1. Question, scope and participation

- Process question:
- Sprint identifier, start and cutoff timestamps:
- Timezone; calendar or business-hour definition:
- Sprint Goal and supporting outcome evidence:
- Repository / board / filter / parent-or-subtask level:
- Applicable completion checklist and version:
- Allowed sources, custodian and dated snapshot:
- Purpose, audience, access, attribution and retention rules:
- Correction / withdrawal process and downstream correction owner:
- Sources excluded from AI input:
- Facilitator and actual accountable owner:
- Required privacy, employment and security reviews:

## 2. Reconcile work and quality

| Item ID | Initially selected? | Added / removed event and time | Retained at cutoff? | Board state | Completion checks and evidence | Goal relationship | Unresolved issue |
| --- | --- | --- | --- | --- | --- | --- | --- |
| | | | | | | | |

- Initial population and count:
- Additions and removals, including reentries:
- Ending population and reconciled count:
- Original-forecast Done numerator / original population:
- Ending-scope Done numerator / ending population:
- Separate goal assessment and reviewer:
- Data errors that prevent calculation:

Do not use an added item in an original-forecast numerator. Preserve event histories when items leave and return. A board label is not automatically completion evidence.

## 3. Build reviewable themes

| Finding ID | Exact narrow statement | Observation / perspective / hypothesis / proposal | Supporting IDs | Counterexamples or disagreement | Missing context | Team correction / status |
| --- | --- | --- | --- | --- | --- | --- |
| | | | | | | |

- Note IDs and revisions available for analysis:
- Duplicate export rule (stable ID + revision, not similar wording):
- Withdrawn or restricted content excluded:
- Number of unique notes; respondent count only if independently established and appropriate:
- Vague statements requiring optional clarification:
- Claims rejected because evidence is insufficient:

No emotion, intent, individual-effort or performance inference from tone or counts. A frequently repeated view is not automatically a verified cause.

## 4. Specify one measurable experiment

- Action ID and approved date:
- Supported process problem:
- Competing explanations still possible:
- Hypothesis:
- Actual named owner and collaborators:
- Change, start, finish and resources:
- Inclusion / exclusion rules and work class:
- Baseline source, value and sample size:
- Numerator, denominator and exact threshold:
- First request / repeated episodes / withdrawal handling:
- Automated acknowledgment versus substantive response definition:
- Cutoff and minimum follow-up:
- Treatment of open, immature and missing-data observations:
- Conditional medians or other supplemental measures:
- Quality guardrail, observation window and source:
- Workload guardrail and source:
- Target signal; adapt / stop conditions:
- Review date and decision owner:

## 5. Record the follow-up without changing the rules

| Request ID | Eligible? Why? | Request timestamp | First substantive response timestamp | Window mature? | Open age at cutoff | Within threshold? | Source / correction |
| --- | --- | --- | --- | --- | --- | --- | --- |
| | | | | | | | |

- Timely responses / all eligible mature requests:
- Immature requests shown separately:
- Unresolved requests and ages:
- Completed-response-only median and sample size, if useful:
- Baseline and follow-up definitions identical? If not, explain and recompute:
- Quality guardrail result; observations not yet mature:
- Workload guardrail result:
- Confounders, changed work mix, staffing and calendar:

Never substitute open age for a completed duration. Do not call an empty export zero delay or zero defects. Define how missing data affects validity.

## 6. Make and revisit the decision

- Adopt / adapt / stop / defer pending required evidence:
- Evidence that supports the decision:
- Guardrail failures or unresolved conditions:
- Team disagreement retained:
- Next action, actual owner and due date:
- Next review / quality maturity date:
- Where the approved decision is recorded:
- Downstream summaries corrected:
- If AI was evaluated: tool, model, configuration, prompt/input version, output, reviewer corrections and permitted retention:

A completed action is not evidence of an effective action. Keep the record small enough for another team member to recompute and challenge.